Skip to content
Development
Skill

/context-engineering

Use when a long session has accumulated stale or conflicting context or the user asks to refresh it. Not for remote, credential, publish, deploy, or irreversible changes.

From plugin
odin-claude-plugin
36200 skills
Install
$ npx -y skills add OutlineDriven/odin-claude-plugin --skill context-engineering --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/context-engineering

Context preview

The summary Claude sees to decide when to auto-load this skill.

Use when a long session has accumulated stale or conflicting context or the user asks to refresh it. Not for remote, credential, publish, deploy, or irreversible changes.

SKILL.md

context-engineering.SKILL.md
name: context-engineering
description: 'Use when a long session has accumulated stale or conflicting context or the user asks to refresh it. Not for remote, credential, publish, deploy, or irreversible changes.'
disable-model-invocation: true

Context engineering

Contract

| Field | Bound contract | |---|---| | Trigger | Long session has accumulated stale or conflicting context, or the user asks to refresh or recheck context before continuing. | | Authority | Reversible local: writes only an optional scratch state note; rollback is undo. No remote mutation. | | Side effect | Optional scratch state note only; no project files touched. | | Done | Acting context is provably current and minimal; stale items are dropped with the user informed. |

Inputs

  • The current acting context present in the session: stated facts, assumptions, inferred file state, and prior decisions.
  • The source of truth for each item being rechecked: repository files, tool output, or user-stated ground truth.
  • Optional: a user-named subset to recheck. If omitted, recheck the full acting context.

Procedure

1. Enumerate every item in the acting context: stated facts, assumptions, inferred file state, and prior decisions. Done when: every acting-context item is enumerated. 2. For each item, recheck it against its source of truth by the narrowest action that settles it: read the file, re-run the tool, or ask the user. Classify the item as current, stale, or contradicted. Done when: every item is classified current, stale, or contradicted. 3. Drop stale and contradicted items from the acting context. Replace a dropped item with the corrected value only when the source of truth supplies one. Done when: stale and contradicted items are dropped and replaced where a corrected value exists. 4. Remove redundant or duplicate items so the acting context holds only items that are provably current and needed for the remaining work. Done when: no redundant or duplicate items remain. 5. Tell the user which items were dropped and why, and identify any item that could not be settled because its source of truth was unreachable. Done when: dropped items and reasons are reported and unsettled items are identified. 6. Optionally write a scratch state note recording the settled context and the dropped items. Do not touch any project file. Done when: the scratch note is written or skipped, and no project file is touched.

Failure and recovery

  • Unreachable source of truth: leave the dependent item flagged as unsettled, do not drop or replace it, and report it to the user. Never assert an item is current without a settled source.
  • Contradiction between sources: report both sources and ask the user which to keep. Do not silently pick one.
  • Partial result: deliver the subset that was settled and explicitly list the unsettled remainder. Never present a partial refresh as complete.
  • Non-mutation boundary: if any step would require touching a project file or remote state, stop and report the boundary rather than widening authority.

Output

Refreshed acting context (only provably current, minimal items) → user-facing list of dropped items with reasons → optional scratch state note. Unsettled items listed separately and remain flagged.

Read more
Ships withodin-claude-plugin

Formerly the ODIN Claude Plugin. The repository URL is unchanged. Outline-Driven Development, nicknamed ODIN, is a highly opinionated code-agent skill library: principles-first engineering, surgical editing, and workflow automation, published as installable

Get the whole plugin
Stats
36
Stars
0
Forks
Active
Maintenance
Python
Language
Apache-2.0
License
3d ago
Last commit
10mo ago
Created

Repo: OutlineDriven/odin-claude-plugin

Other skills on odin-claude-plugin.